Short-term load forecasting of electric vehicle charging stations based on adaptive VMD-BiLSTM
Minan Tang, Wenxin Sheng, Jiandong Qiu, Chenchen Zhang, Hanting Li
Abstract
The forecasting of electric vehicle charging station loads is crucial for ensuring timely and feasible energy supply to achieve satisfactory charging for the relevant vehicles. Aiming at problems such as low accuracy of short-term prediction of electric vehicle charging piles, a short-term prediction method for charging a load of electric vehicle charging stations based on variational modal decomposition (VMD)-frost and ice optimisation algorithm (RIME)-bi-directional long and short term memory neural network (BiLSTM) model is proposed. Firs, in order to weaken the prediction difficulties caused by information redundancy, the optimal number of decomposition layers and penalty factors are found by RIME optimisation VMD, and the load data are decomposed into smooth IMFs as inputs. Second, in order to overcome the problem of poor neural network parameters, different components are fed into the BiLSTM prediction model optimised by the RIME algorithm to make the model robust. Finally, the public dataset of ANN-DATA in the United States and the historical load data of a charging station in Lanzhou, China are validated, and a comparison of load forecasts as well as an error analysis are performed. The results show that the model outperforms the single model and other algorithmic optimisation models in all three performance metrics on all datasets, with smaller errors and higher prediction accuracy.